From 35e2088b7ea9e7c299bcb29b6fa6ae98fd2ba566 Mon Sep 17 00:00:00 2001
From: Pere Diaz Bou <pere-altea@hotmail.com>
Date: Mon, 19 May 2025 15:00:32 +0200
Subject: [PATCH] cacheflush move dirty page to new snapshot

After inserting a page into the wal, we dispose of the modified page.
This is unnecessary as we can simply move new page to the newest
snapshot where this page can be read.
---
 core/storage/pager.rs | 14 ++++++++++++++
 1 file changed, 14 insertions(+)

diff --git a/core/storage/pager.rs b/core/storage/pager.rs
index 7d1acde7d..f369c9d57 100644
--- a/core/storage/pager.rs
+++ b/core/storage/pager.rs
@@ -461,6 +461,9 @@ impl Pager {
                         Some(wal) => wal.borrow().get_max_frame(),
                         None => 0,
                     };
+                    let max_frame_after_append = self.wal.as_ref().map(|wal| {
+                        wal.borrow().get_max_frame() + self.dirty_pages.borrow().len() as u64
+                    });
                     for page_id in self.dirty_pages.borrow().iter() {
                         let mut cache = self.page_cache.write();
                         let page_key = PageCacheKey::new(*page_id, Some(max_frame));
@@ -473,6 +476,17 @@ impl Pager {
                                 db_size,
                                 self.flush_info.borrow().in_flight_writes.clone(),
                             )?;
+                            // Assuming writer will always end append frames at frameid == this_transaction.max_frame + dirty_pages,
+                            // we can insert this page with a new key into that new "snapshot" that has the newest max frame. We don't
+                            // simply clone the page and insert with new key because next cache.delete will invalidate the contents of the page,
+                            // therefore we need to clone the contents itself and place them on the new page. Cloning contents should be fast because
+                            // buffer is wrapped around an Arc.
+                            let new_page = Page::new(*page_id);
+                            new_page.get().contents = Some(page.get_contents().clone());
+                            new_page.set_loaded();
+                            let new_page: Arc<Page> = Arc::new(new_page);
+                            let new_page_key = PageCacheKey::new(*page_id, max_frame_after_append);
+                            cache.insert(new_page_key, new_page).map_err(|e| {LimboError::InternalError(format!("Failed to delete page {:?} from cache during flush: {:?}. Might be actively referenced.", page_id, e))})?;
                         }
                         page.clear_dirty();
                         // This page is no longer valid.